1. Introduction
Basalt is a natural material that is found in volcanic rocks originating from frozen lava, with a melting temperature of 1500–1700 °C. The basalt fiber (BF) is produced by heating basalt rock in the furnace at 1450 to 1500 °C. The molten material is then passed through platinum and rhodium crucible bushing to form fiber. This technology is known as continuous spinning. The development of BF was first performed by the Moscow Research Institute of Glass and Plastic in 1953–1954, and its first industrial production furnace was completed in 1985 at a Ukraine fiber laboratory [
1]. The fibers of basalt provide resistance against corrosion, heat, and chemical attack on the concrete, making them beneficial for use in building materials [
2,
3,
4,
5,
6,
7]. Previous studies have shown that BF is able to enhance the mechanical characteristics of concrete [
8,
9,
10]. For example, Kharun et al. [
11] found that the durability and stability of high-strength concrete (HSC) was increased after adding different percentages of BF. They found that BF improves flexural strength and prevents development of cracking in concrete blocks. Their results demonstrated that the optimum percentage of BF added to HSC is 1% (by concrete volume). With the addition of 1% BF, the compressive and tensile strengths of HSC increased more than 37 and 70%, respectively. Yang et al. [
12] indicated that the addition of a proper amount of BF to ordinary concrete can delay initial cracking and increase the toughness of concrete. Long cracks on the surface of concrete gradually transform into many micro-cracks with increasing fiber content. They also reported that adding 0.6% BF can increase the compressive strength of concrete more than 13%. Biradar et al. [
13] found that concrete containing 0.3% BF illustrated maximum strength values. They showed that the compressive, tensile, and flexural strength of concrete reinforced with 0.3% BF was 9.82, 36.7, and 18.83% more than their corresponding values in ordinary concrete, respectively. Katkhuda and Shatarat [
14] reported that chopped BF had insignificant effects on the increase in compressive strength but considerably enhanced flexural and tensile strengths of recycled concrete aggregate.
High-performance concrete (HPC) has become very popular due durability and resistance against penetration of aggressive agents. HPC is widely used in structures such as tall buildings, bridges, runways, and highway pavements [
15]. However, HPC is still considered a brittle material with poor tensile and flexural strengths. Therefore, it would be interesting to study the effect of BF on the mechanical properties of HPC. Ayub et al. [
16,
17] estimated the impact of chopped BF on the mechanical properties of HPC. Their results showed that adding 2% BF (by volume of the concrete) improved compressive strength more than 17%. They also mentioned the effect of BF on the enhancement of HPC ductility. Mohaghegh et al. [
18] studied the impact of different percentages of BF on the characteristics of HPC and found that the highest flexural, tensile, and compressive strength values in HPC were achieved with the addition of 1.33% BF. According to their results, adding BF may be an effective way for enhancing the mechanical characteristics of HPC. Kharun et al. [
19] investigated the effect of adding chopped BF on HPC characteristics. Their results proved that adding 1.2% BF had maximum effects on the improvement of mechanical characteristics. Thus, based on the results of previous investigations [
16,
17,
18,
19,
20], the optimum percentage of BF to be added to HPC is usually more than 1% (by concrete volume).
On the other hand, machine; earning (ML), as a sub-field of artificial intelligence, has attracted attention in various industries, such as the medical, mechanical, and manufacturing industries [
21,
22,
23]. ML has been greatly applied to the material sciences and civil engineering within the past decade, mainly for the prediction of the mechanical properties of different types of concretes [
24,
25,
26,
27,
28,
29,
30,
31,
32,
33,
34,
35,
36]. Due to the effectiveness and capabilities of the modern computational processes, ML techniques can assess the mechanical properties of concrete without spending time in the laboratory, or investing money in experimentation [
37]. For example, Su et al. [
38] applied multiple linear regression (MLR), support vector machine (SVM), and artificial neural network (ANN) methods to forecast the mechanical characteristics of reinforced concrete with polymer fibers. They found that the developed SVM algorithm presented the best prediction results. In another example, Nguyen et al. [
39] employed ML to find the best method for prediction of the compressive strength of geo-polymer reinforced concrete, out of 335 mix proportions. Sami Ullah et al. [
40] applied support vector regression (SVR) and random forest (RF) techniques for strength assessment of lightweight foam concrete (LFC). They depicted that acceptable accuracy in the prediction of compressive strength of LFC was obtained by employing the RF. According to their findings, the coefficient of determination (R
2) value was 0.95 via RF. They concluded that the RF method was able to predict compressive strength with high accuracy. Liu [
41] investigated the prediction of the mechanical properties of HPC using extreme gradient boosting (XGBoost), SVR, and RF. He showed that the XGBoost algorithm has appropriate performance in predicting the compressive strength of HPC. The R
2 > 0.99 was indicative of the high accuracy of his model in the prediction. Kashyzadeh et al. [
42] utilized a back-propagation neural network (BPNN) optimized by a genetic algorithm (GA) to predict the compressive strength of concrete. Their objective was to find the predictive results of concrete strength via analyzing the curing temperature and the shape and size of aggregates as the most important variables. They illustrated that the developed neural network methods were consistent with the experimental ones.
Although researchers prefer to predict the mechanical properties of concrete through ML, their research fields have mainly concentrated on predicting some limited characteristics, such as compressive strength. They did not consider other mechanical properties, such as flexural and tensile strengths, which play important roles in concrete behavior. In the present paper, not only the compressive strength, but also the flexural and tensile strengths of concrete were predicted through three ML techniques, including linear regression (LR), SVR, and polynomial regression (PR). Then, compressive stress–strain curves were simulated through PR. Moreover, the values of the modulus of elasticity (ME) were estimated using ML techniques and compared with the relations existing in the literature. Finally, some correlations were proposed between the compressive, flexural, and tensile strengths.
4. Conclusions
In this research, ML techniques were applied for the prediction of the different mechanical properties of BFHPC. For this purpose, three predictive algorithms, LR, SVR, and PR, were employed and analyzed to find the most accurate predictions. Moreover, the compressive stress–strain curves of samples were simulated by fitting through the prediction points. The prediction of the modulus of elasticity (ME) and comparison of forecasted ME with the relations available in the literature was also part of this research. Finally, some correlations between compressive, tensile, and flexural strengths of BFHPC were suggested. The following results can be drawn from this investigation:
The mechanical characteristics of BFHPC can be more accurately predicted via PR in comparison with LR and SVR. For example, in predicting the compressive strength through PR, the values of R2, RMSE, and MAE were 0.99, 0.05 Mpa, and 0.19 MPa, respectively. This confirms the high accuracy of PR in terms of its prediction.
Although simulation of compressive stress–strain curves has challenges (particularly simulation of the plastic phase), the PR technique was able to appropriately forecast these curves.
The predicted values of ME, one the most important properties of concrete, using PR were close to the experimental results and results of some available formulas in the literature.
Proposed models could be efficiently used at the construction site to minimize required laboratory work, as well as save time and costs.
More reliable and high-quality experimental data will play a vital role in improving the model performance. In other words, a database and more input parameters may be required for generation of a better response from employed models in future. ML techniques may be utilized with heuristic approaches, such as the whale optimization algorithm, ant colony optimization, and particle swarm optimization, to achieve more effective and precise outcomes. Future studies should compare these tactics with current findings.